MUMBAI, India, June 22 -- Intellectual Property India has published a patent application (202641068705 A) filed by Pes University on June 01, 2026, for System And Method For Detecting Anomalies In Media Acquisition Data.
Inventors include Ankush S; Aniruddh Suresh Guttikar; Aditya B; Arun Kumar R; and Chitra G M.
The application for the patent was published on June 12, 2026, under issue no. 24/2026.
Abstract: The present disclosure relates to a system (102) and method (1100) for detecting anomalies in media acquisition data. The system employs a pre-trained and frozen spatiotemporal feature extraction model to generate semantic representations from sequences of media frames, ensuring computational efficiency and prevention of catastrophic forgetting. The generated feature representation is processed through a dual-pathway architecture comprising a specialist pathway for detecting predefined anomaly classes using supervised learning, and a generalist pathway for identifying unknown anomalies through zero-shot semantic similarity with text-based representations. A hierarchical arbitration logic determines the anomaly state by prioritizing high-confidence outputs from the specialist pathway while selectively validating generalist outputs to reduce false positives. Further, temporal persistence validation ensures robustness by confirming anomalies across consecutive frames. The system outputs validated anomaly results with classification, confidence, and temporal metadata, enabling reliable real-time monitoring and forensic analysis.
Disclaimer: Curated by HT Syndication.
